A field test designed to measure atomospheric refraction error in historical and modern leveling was conducted in May-June 1981 on a 50-km-long grade from Saugus to Palmdale, California. During 1955-1971, the length of sights made between the level instrument and rods systematically decreased from 60 m to 26 m. The difference in height near Palmdale measured by single-run long-sight (42-m) and short-sight (22-m) leveling during the test was 6 times larger than expected random error. Correcction for refraction by using either the observed or modeled vertical temperature gradient in K¿kkamaki's balanced sight equation reduced the height difference to the level of random error uncertainty. The observed temperature gradient obeyed a power law relation, T =a+bzc, where z is the height and a, b, and c are constants that depended on atmospheric conditions and the ground surface beneath the line of sight. The refraction-corrected leveling satisfies the specifications and meets all standards of first-order control surveys. The six historical surveys of the Saugus-Palmdale grade were corrected for refraction error using the results of the experimental and for rod scale errors and nontectonic subsidence consisted in previous investigations. The corrected uplift near Palmdale reached 56¿16 mm with respect to Saugus during this period 1955--1965. This amount of uplift is about one third that obtained before removal of refraction error. The corrected displacement profiles also reveal previously unrecognized deformation in the epicentral region of the 1971 San Fernando ML = 6.4 earthquake during the decade before the main shock.